Mistral’s new model, Mistral Large 4—nicknamed “Le Chonk”—is available as a public preview API, but its weights were not yet downloadable at launch on October 6, 2026. Mistral promotes it as a leading open-weight model outside China; that ranking is the company’s claim, not an independently established comparison across named Chinese competitors and shared benchmarks.
What is Le Chonk?
“Le Chonk” is the nickname for Mistral Large 4, announced by French AI company Mistral on October 6, 2026. Mistral calls it a natively multimodal model with 1 trillion total parameters and 49 billion active parameters. It also says the model was trained on 3,800 NVIDIA Grace Blackwell GPUs in its European datacenters. These are company-reported specifications. In its launch post, Mistral wrote: “Today, we’re launching a public preview of Mistral Large 4. Unofficially ML4, very officially: le Chonk.” Mistral’s announcement
As an Amazon Associate I earn from qualifying purchases.
“Open-weight” describes the planned release of model weights; it should not be taken to mean “open-source.” The launch information did not establish the final license or all conditions for using and modifying those weights.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCan you use or download Mistral Large 4 now?
Preview API: available at launch
Mistral said the public preview API was available through Mistral Studio. Its launch announcement listed a price of $1.36 per million input tokens and $4.18 per million output tokens. These are Mistral’s preview API prices as stated on October 6, 2026, and may change. Mistral’s announcement
#1 Best Overall
Weights: announced, not yet released at launch
Mistral said the weights would arrive by the end of October. Reuters reported October 27 as the public release date. At the October 6 announcement, that release was still pending; check Mistral’s current product information before relying on the date. Mistral said the release would add architecture details, benchmarks, and post-training methodology. The launch sources did not settle the final license or the hardware required to run the model yourself. Mistral’s announcement; Reuters, republished by Euronext
What performance has Mistral reported?
The figures below come from Mistral’s October 6, 2026 launch announcement. They are company-reported results, not independent confirmation, and scores across different tasks should not be treated as directly interchangeable.
| Area | Mistral-reported result |
|---|---|
| Cybersecurity | 82% on a test involving reproducing a real vulnerability in open-source software and patching it; Mistral described this as the highest score on that test. It also reported 93% of challenges on Cybench, which it described as 40 security exercises. |
| Coding and software agents | 61.7% on DeepSWE v1.1; 59.4% on SWE-Atlas-QnA; 28.3% on Terminal-Bench 4; and 49.8% on the combined Coding Agent Index. Mistral said the latter was ahead of DeepSeek V4 Pro 0813. |
| Blind coding-quality assessment | 3.74 out of 5 from professional annotators evaluating coding quality with model identities hidden; Mistral said the preview placed second among five models. |
| Business workflows | 59.9% on AutomationBench, which Mistral described as covering 657 workflows. |
| Finance | 1,393 Elo on AA-Briefcase. |
| Visual grounding | 42% for Large 4 versus 41% for GPT-6-Astra on Dense 200. |
Mistral also highlights agentic workflows and performance in finance, law, and manufacturing. Its announcement makes broader state-of-the-art claims in some of these areas, but the cited launch coverage does not independently validate every claim. A benchmark result is most useful when the task, scoring method, and evaluator are clear; it does not by itself establish that a model will perform best on a particular company’s workload. Mistral’s announcement; WIRED’s launch coverage
Is Le Chonk really the best open-weight model outside China?
That wording is Mistral’s positioning, not a settled independent ranking. Mistral CEO Artur Mensch told Reuters that the model was above Chinese models “on certain aspects, including cyber.” Reuters reported that he did not identify which Chinese models or benchmarks he meant. Without those details and a common evaluation, the statement cannot establish a general win over Chinese open-weight models. Reuters, republished by Euronext
Mistral cofounder and chief scientist Guillaume Lample told WIRED, “There are a lot of areas where the other labs will not focus that much,” and said, “Mistral is still in the race of getting the best model.” Those comments describe the company’s strategy and ambition; they are not an independent comparison. WIRED’s launch coverage
For a useful comparison, look at the same named benchmark and task, whether the evaluation is independent or developer-run, and the model’s performance on your own workload. Also compare access method, weight license, deployment requirements, and the control or regional governance you need. At launch, the license and self-hosting hardware requirements for Large 4 were not established.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Who should pay attention to the launch?
Developers who want to try the model can evaluate the preview API in Mistral Studio, taking its token pricing into account. Teams considering self-hosting should wait for the weights and verify the release license and hardware requirements before planning deployment. Mistral also describes Mistral Forge as a route for enterprise training and customization, but the launch announcement does not make that a substitute for evaluating Large 4 on a team’s specific tasks. Mistral’s announcement
Recommended Free Tools
The launch is notable for the combination of a large, multimodal model and Mistral’s reported coding, security, business-workflow, finance, and visual-grounding results. The distinction to keep in view is what can be tried now—the preview API—and what was still forthcoming at announcement—the downloadable weights and the details needed to assess self-hosting and licensing.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
